{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:KQUOS7KNFUMFRPOTQVYTYNK3IT","short_pith_number":"pith:KQUOS7KN","schema_version":"1.0","canonical_sha256":"5428e97d4d2d1858bdd385713c355b44f3f6a78f0aebf6cb297019eb2d2422c1","source":{"kind":"arxiv","id":"2506.02678","version":3},"attestation_state":"computed","paper":{"title":"TL;DR: Too Long, Do Re-weighting for Efficient LLM Reasoning Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CE","cs.NA","math.NA"],"primary_cat":"cs.CL","authors_text":"Cheng-Lin Liu, Fei Yin, Haizhen Huang, Haotian Xu, Lei Ji, Peijie Wang, Weiwei Deng, Xiao Liang, Xiao Liu, Xing W, Yeyun Gong, Zhijiang Guo, Zhong-Zhi Li, Zihao Tang","submitted_at":"2025-06-03T09:23:41Z","abstract_excerpt":"Large Language Models (LLMs) have recently achieved remarkable progress by leveraging Reinforcement Learning and extended Chain-of-Thought (CoT) techniques. However, the challenge of performing efficient language reasoning--especially during inference with extremely long outputs--has drawn increasing attention from the research community. In this work, we propose a dynamic ratio-based training pipeline that does not rely on sophisticated data annotations or interpolation between multiple models. We continuously balance the weights between the model's System-1 and System-2 data to eliminate red"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2506.02678","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-03T09:23:41Z","cross_cats_sorted":["cs.CE","cs.NA","math.NA"],"title_canon_sha256":"c441a693620553046afcb5f427dd1ca8ac93a4f9369d9e9854e7297735aefc1d","abstract_canon_sha256":"a61f6891ef4484f55444ed561b3022098d0d22e4d489f8e414ba9aafb464ff87"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:21:48.005280Z","signature_b64":"47XtvHqPmREV2za9bZJHAr9NiZPJxqLh2bYyDR8mrthg0h0qOpukxgwA520WrVUu5CSZP1CTEcgh16oz3A+eAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5428e97d4d2d1858bdd385713c355b44f3f6a78f0aebf6cb297019eb2d2422c1","last_reissued_at":"2026-07-05T11:21:48.004779Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:21:48.004779Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TL;DR: Too Long, Do Re-weighting for Efficient LLM Reasoning Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CE","cs.NA","math.NA"],"primary_cat":"cs.CL","authors_text":"Cheng-Lin Liu, Fei Yin, Haizhen Huang, Haotian Xu, Lei Ji, Peijie Wang, Weiwei Deng, Xiao Liang, Xiao Liu, Xing W, Yeyun Gong, Zhijiang Guo, Zhong-Zhi Li, Zihao Tang","submitted_at":"2025-06-03T09:23:41Z","abstract_excerpt":"Large Language Models (LLMs) have recently achieved remarkable progress by leveraging Reinforcement Learning and extended Chain-of-Thought (CoT) techniques. However, the challenge of performing efficient language reasoning--especially during inference with extremely long outputs--has drawn increasing attention from the research community. In this work, we propose a dynamic ratio-based training pipeline that does not rely on sophisticated data annotations or interpolation between multiple models. We continuously balance the weights between the model's System-1 and System-2 data to eliminate red"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.02678","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2506.02678/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2506.02678","created_at":"2026-07-05T11:21:48.004840+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.02678v3","created_at":"2026-07-05T11:21:48.004840+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.02678","created_at":"2026-07-05T11:21:48.004840+00:00"},{"alias_kind":"pith_short_12","alias_value":"KQUOS7KNFUMF","created_at":"2026-07-05T11:21:48.004840+00:00"},{"alias_kind":"pith_short_16","alias_value":"KQUOS7KNFUMFRPOT","created_at":"2026-07-05T11:21:48.004840+00:00"},{"alias_kind":"pith_short_8","alias_value":"KQUOS7KN","created_at":"2026-07-05T11:21:48.004840+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11052","citing_title":"Attention Amnesia in Hybrid LLMs: When CoT Fine-Tuning Breaks Long-Range Recall, and How to Fix It","ref_index":80,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17672","citing_title":"Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2509.02547","citing_title":"The Landscape of Agentic Reinforcement Learning for LLMs: A Survey","ref_index":210,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT","json":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT.json","graph_json":"https://pith.science/api/pith-number/KQUOS7KNFUMFRPOTQVYTYNK3IT/graph.json","events_json":"https://pith.science/api/pith-number/KQUOS7KNFUMFRPOTQVYTYNK3IT/events.json","paper":"https://pith.science/paper/KQUOS7KN"},"agent_actions":{"view_html":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT","download_json":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT.json","view_paper":"https://pith.science/paper/KQUOS7KN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.02678&json=true","fetch_graph":"https://pith.science/api/pith-number/KQUOS7KNFUMFRPOTQVYTYNK3IT/graph.json","fetch_events":"https://pith.science/api/pith-number/KQUOS7KNFUMFRPOTQVYTYNK3IT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT/action/storage_attestation","attest_author":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT/action/author_attestation","sign_citation":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT/action/citation_signature","submit_replication":"https://pith.science/pith/KQUOS7KNFUMFRPOTQVYTYNK3IT/action/replication_record"}},"created_at":"2026-07-05T11:21:48.004840+00:00","updated_at":"2026-07-05T11:21:48.004840+00:00"}